def on_build(self, in_ch, d_ch, d_mask_ch, **kwargs):
self.upscale0 = Upscale(in_ch, d_ch * 8, kernel_size=3)
self.upscale1 = Upscale(d_ch * 8, d_ch * 4, kernel_size=3)
self.upscale2 = Upscale(d_ch * 4, d_ch * 2, kernel_size=3)
self.res0 = ResidualBlock(d_ch * 8, kernel_size=3)
self.res1 = ResidualBlock(d_ch * 4, kernel_size=3)
self.res2 = ResidualBlock(d_ch * 2, kernel_size=3)
self.out_conv = nn.Conv2D(d_ch * 2,
3,
kernel_size=1,
padding='SAME')
self.upscalem0 = Upscale(in_ch,
d_mask_ch * 8,
kernel_size=3)
self.upscalem1 = Upscale(d_mask_ch * 8,
d_mask_ch * 4,
kernel_size=3)
self.upscalem2 = Upscale(d_mask_ch * 4,
d_mask_ch * 2,
kernel_size=3)
self.out_convm = nn.Conv2D(d_mask_ch * 2,
1,
kernel_size=1,
padding='SAME')
def on_build(self, patch_size, in_ch, base_ch = 16, use_fp16 = False):
self.use_fp16 = use_fp16
conv_dtype = tf.float16 if use_fp16 else tf.float32
class ResidualBlock(nn.ModelBase):
def on_build(self, ch, kernel_size=3 ):
self.conv1 = nn.Conv2D( ch, ch, kernel_size=kernel_size, padding='SAME', dtype=conv_dtype)
self.conv2 = nn.Conv2D( ch, ch, kernel_size=kernel_size, padding='SAME', dtype=conv_dtype)
def forward(self, inp):
x = self.conv1(inp)
x = tf.nn.leaky_relu(x, 0.2)
x = self.conv2(x)
x = tf.nn.leaky_relu(inp + x, 0.2)
return x
prev_ch = in_ch
self.convs = []
self.upconvs = []
layers = self.find_archi(patch_size)
level_chs = { i-1:v for i,v in enumerate([ min( base_ch * (2**i), 512 ) for i in range(len(layers)+1)]) }
self.in_conv = nn.Conv2D( in_ch, level_chs[-1], kernel_size=1, padding='VALID', dtype=conv_dtype)
for i, (kernel_size, strides) in enumerate(layers):
self.convs.append ( nn.Conv2D( level_chs[i-1], level_chs[i], kernel_size=kernel_size, strides=strides, padding='SAME', dtype=conv_dtype) )
self.upconvs.insert (0, nn.Conv2DTranspose( level_chs[i]*(2 if i != len(layers)-1 else 1), level_chs[i-1], kernel_size=kernel_size, strides=strides, padding='SAME', dtype=conv_dtype) )
self.out_conv = nn.Conv2D( level_chs[-1]*2, 1, kernel_size=1, padding='VALID', dtype=conv_dtype)
self.center_out = nn.Conv2D( level_chs[len(layers)-1], 1, kernel_size=1, padding='VALID', dtype=conv_dtype)
self.center_conv = nn.Conv2D( level_chs[len(layers)-1], level_chs[len(layers)-1], kernel_size=1, padding='VALID', dtype=conv_dtype)
def on_build(self, in_ch, d_ch, d_mask_ch):
self.upscale0 = Upscale(in_ch, d_ch * 8, kernel_size=3)
self.upscale1 = Upscale(d_ch * 8, d_ch * 4, kernel_size=3)
self.upscale2 = Upscale(d_ch * 4, d_ch * 2, kernel_size=3)
self.upscale3 = Upscale(d_ch * 2, d_ch * 1, kernel_size=3)
self.res0 = ResidualBlock(d_ch * 8, kernel_size=3)
self.res1 = ResidualBlock(d_ch * 4, kernel_size=3)
self.res2 = ResidualBlock(d_ch * 2, kernel_size=3)
self.res3 = ResidualBlock(d_ch * 1, kernel_size=3)
self.out_conv = nn.Conv2D(
d_ch * 1,
3,
kernel_size=1,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
self.upscalem0 = Upscale(in_ch, d_mask_ch * 8, kernel_size=3)
self.upscalem1 = Upscale(d_mask_ch * 8,
d_mask_ch * 4,
kernel_size=3)
self.upscalem2 = Upscale(d_mask_ch * 4,
d_mask_ch * 2,
kernel_size=3)
self.upscalem3 = Upscale(d_mask_ch * 2,
d_mask_ch * 1,
kernel_size=3)
self.out_convm = nn.Conv2D(
d_mask_ch * 1,
1,
kernel_size=1,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
def on_build(self):
self.conv1 = nn.Conv2D (3, 64, kernel_size=7, strides=2, padding='SAME')
self.bn1 = nn.BatchNorm2D(64)
self.conv2 = ConvBlock(64, 128)
self.conv3 = ConvBlock(128, 128)
self.conv4 = ConvBlock(128, 256)
self.m = []
self.top_m = []
self.conv_last = []
self.bn_end = []
self.l = []
self.bl = []
self.al = []
for i in range(4):
self.m += [ HourGlass(256, 4) ]
self.top_m += [ ConvBlock(256, 256) ]
self.conv_last += [ nn.Conv2D (256, 256, kernel_size=1, strides=1, padding='VALID') ]
self.bn_end += [ nn.BatchNorm2D(256) ]
self.l += [ nn.Conv2D (256, 68, kernel_size=1, strides=1, padding='VALID') ]
if i < 4-1:
self.bl += [ nn.Conv2D (256, 256, kernel_size=1, strides=1, padding='VALID') ]
self.al += [ nn.Conv2D (68, 256, kernel_size=1, strides=1, padding='VALID') ]
def on_build(self, in_ch, d_ch):
self.upscale0_1 = Upscale(in_ch, d_ch * 1)
self.upscale0_2 = Upscale(d_ch * 1, d_ch * 1)
self.upscale0_3 = Upscale(d_ch * 1, d_ch * 1)
self.out0 = nn.Conv2D(
d_ch * 1,
3,
kernel_size=1,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
self.upscale1_1 = Upscale(in_ch, d_ch * 1)
self.upscale1_2 = Upscale(d_ch * 2, d_ch * 1)
self.upscale1_3 = Upscale(d_ch * 2, d_ch * 1)
self.out1 = nn.Conv2D(
d_ch * 2,
3,
kernel_size=1,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
self.upscalem1 = Upscale(in_ch, d_ch)
self.upscalem2 = Upscale(d_ch, d_ch // 2)
self.upscalem3 = Upscale(d_ch // 2, d_ch // 2)
self.outm = nn.Conv2D(
d_ch // 2,
1,
kernel_size=1,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
def on_build(self, in_ch):
self.upscale6 = Upscale(in_ch, in_ch//2)
self.upscale5 = Upscale(in_ch//2, in_ch//4)
self.upscale4 = Upscale(in_ch//4, in_ch//8)
self.upscale3 = Upscale(in_ch//8, in_ch//16)
self.upscale2 = Upscale(in_ch//16, in_ch//32)
#self.upscale1 = Upscale(in_ch//32, in_ch//64)
self.out_conv = nn.Conv2D( in_ch//32, 3, kernel_size=1, padding='SAME', kernel_initializer=conv_kernel_initializer)
self.res61 = ResidualBlock(in_ch//2, mod=8)
self.res62 = ResidualBlock(in_ch//2, mod=8)
self.res63 = ResidualBlock(in_ch//2, mod=8)
self.res51 = ResidualBlock(in_ch//4, mod=8)
self.res52 = ResidualBlock(in_ch//4, mod=8)
self.res53 = ResidualBlock(in_ch//4, mod=8)
self.res41 = ResidualBlock(in_ch//8, mod=8)
self.res42 = ResidualBlock(in_ch//8, mod=8)
self.res43 = ResidualBlock(in_ch//8, mod=8)
self.res31 = ResidualBlock(in_ch//16, mod=8)
self.res32 = ResidualBlock(in_ch//16, mod=8)
self.res33 = ResidualBlock(in_ch//16, mod=8)
self.res21 = ResidualBlock(in_ch//32, mod=8)
self.res22 = ResidualBlock(in_ch//32, mod=8)
self.res23 = ResidualBlock(in_ch//32, mod=8)
m_ch = in_ch//2
self.upscalem6 = Upscale(in_ch, m_ch//2)
self.upscalem5 = Upscale(m_ch//2, m_ch//4)
self.upscalem4 = Upscale(m_ch//4, m_ch//8)
self.upscalem3 = Upscale(m_ch//8, m_ch//16)
self.upscalem2 = Upscale(m_ch//16, m_ch//32)
#self.upscalem1 = Upscale(m_ch//32, m_ch//64)
self.out_convm = nn.Conv2D( m_ch//32, 1, kernel_size=1, padding='SAME', kernel_initializer=conv_kernel_initializer)
def on_build(self,
patch_size,
in_ch,
base_ch=None,
conv_kernel_initializer=None):
suggested_base_ch, kernels_strides = patch_discriminator_kernels[
patch_size]
if base_ch is None:
base_ch = suggested_base_ch
prev_ch = in_ch
self.convs = []
for i, (kernel_size, strides) in enumerate(kernels_strides):
cur_ch = base_ch * min((2**i), 8)
self.convs.append(
nn.Conv2D(prev_ch,
cur_ch,
kernel_size=kernel_size,
strides=strides,
padding='SAME',
kernel_initializer=conv_kernel_initializer))
prev_ch = cur_ch
self.out_conv = nn.Conv2D(prev_ch,
1,
kernel_size=1,
padding='VALID',
kernel_initializer=conv_kernel_initializer)
def on_build(self):
in_ch, ae_ch = self.in_ch, self.ae_ch
self.dense_conv1 = nn.Conv2D( in_ch, 64, kernel_size=1, padding='SAME', kernel_initializer=conv_kernel_initializer)
self.dense_conv2 = nn.Conv2D( 64, in_ch, kernel_size=1, padding='SAME', kernel_initializer=conv_kernel_initializer)
self.conv7 = Upscale(in_ch, in_ch//2)
self.conv6 = Upscale(in_ch//2, in_ch//4)
def on_build(self, ch, kernel_size=3):
self.conv1 = nn.Conv2D(ch,
ch,
kernel_size=kernel_size,
padding='SAME')
self.conv2 = nn.Conv2D(ch,
ch,
kernel_size=kernel_size,
padding='SAME')
def on_build(self, ch, kernel_size=3):
self.conv1 = nn.Conv2D(ch,
ch * 2,
kernel_size=kernel_size,
padding='SAME')
self.conv2 = nn.Conv2D(ch * 2,
ch,
kernel_size=kernel_size,
padding='SAME')
self.scale_add = nn.ScaleAdd(ch)
def on_build(self, in_ch, out_ch, level):
self.zero_level = level == 0
self.conv1 = nn.Conv2D(
in_ch,
out_ch,
kernel_size=4 if self.zero_level else 3,
padding=3 if self.zero_level else 'SAME')
self.conv2 = nn.Conv2D(out_ch,
out_ch,
kernel_size=3,
padding='SAME')
def on_build(self, in_ch, code_res, ch=256):
n_downscales = 2 + code_res // 8
self.convs = []
prev_ch = in_ch
for i in range(n_downscales):
cur_ch = ch * min( (2**i), 8 )
self.convs.append ( nn.Conv2D( prev_ch, cur_ch, kernel_size=4 if i == 0 else 3, strides=2, padding='SAME', kernel_initializer=conv_kernel_initializer) )
prev_ch = cur_ch
self.out_conv = nn.Conv2D( prev_ch, 1, kernel_size=1, padding='VALID', kernel_initializer=conv_kernel_initializer)
def on_build(self, ch, kernel_size=3):
self.conv1 = nn.Conv2D(
ch,
ch,
kernel_size=kernel_size,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
self.conv2 = nn.Conv2D(
ch,
ch,
kernel_size=kernel_size,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
def on_build(self, in_ch, d_ch):
self.upscale1 = Upscale(in_ch, d_ch*4)
self.res1 = ResidualBlock(d_ch*4)
self.upscale2 = Upscale(d_ch*4, d_ch*2)
self.res2 = ResidualBlock(d_ch*2)
self.upscale3 = Upscale(d_ch*2, d_ch*1)
self.res3 = ResidualBlock(d_ch*1)
self.upscalem1 = Upscale(in_ch, d_ch)
self.upscalem2 = Upscale(d_ch, d_ch//2)
self.upscalem3 = Upscale(d_ch//2, d_ch//2)
self.out_conv = nn.Conv2D( d_ch*1, 3, kernel_size=1, padding='SAME')
self.out_convm = nn.Conv2D( d_ch//2, 1, kernel_size=1, padding='SAME')
def on_build(self, in_ch, out_ch):
self.conv = nn.Conv2D(in_ch,
out_ch,
kernel_size=3,
padding='SAME')
self.frn = nn.FRNorm2D(out_ch)
self.tlu = nn.TLU(out_ch)
def on_build(self, base_ch, out_ch):
self.up4 = UpConvBlock(base_ch * 8, base_ch * 4)
self.conv43 = ConvBlock(base_ch * 12, base_ch * 8)
self.conv42 = ConvBlock(base_ch * 8, base_ch * 8)
self.conv41 = ConvBlock(base_ch * 8, base_ch * 8)
self.up3 = UpConvBlock(base_ch * 8, base_ch * 4)
self.conv33 = ConvBlock(base_ch * 12, base_ch * 8)
self.conv32 = ConvBlock(base_ch * 8, base_ch * 8)
self.conv31 = ConvBlock(base_ch * 8, base_ch * 8)
self.up2 = UpConvBlock(base_ch * 8, base_ch * 4)
self.conv23 = ConvBlock(base_ch * 8, base_ch * 4)
self.conv22 = ConvBlock(base_ch * 4, base_ch * 4)
self.conv21 = ConvBlock(base_ch * 4, base_ch * 4)
self.up1 = UpConvBlock(base_ch * 4, base_ch * 2)
self.conv12 = ConvBlock(base_ch * 4, base_ch * 2)
self.conv11 = ConvBlock(base_ch * 2, base_ch * 2)
self.up0 = UpConvBlock(base_ch * 2, base_ch)
self.conv02 = ConvBlock(base_ch * 2, base_ch)
self.conv01 = ConvBlock(base_ch, base_ch)
self.out_conv = nn.Conv2D(base_ch,
out_ch,
kernel_size=3,
padding='SAME')
def on_build(self, in_ch, out_ch, kernel_size=3):
self.conv1 = nn.Conv2D(
in_ch,
out_ch * 4,
kernel_size=kernel_size,
padding='SAME',
kernel_initializer=conv_kernel_initializer)
def on_build(self, *args, **kwargs):
self.conv1 = nn.Conv2D(self.in_ch,
self.out_ch,
kernel_size=self.kernel_size,
strides=2,
padding='SAME',
dtype=conv_dtype)
def on_build(self, *args, **kwargs):
self.conv1 = nn.Conv2D(self.in_ch,
self.out_ch //
(4 if self.subpixel else 1),
kernel_size=self.kernel_size,
strides=1 if self.subpixel else 2,
padding='SAME',
dilations=self.dilations)
def on_build(self, in_ch, d_ch, d_mask_ch):
self.upscale0 = Upscale(in_ch, d_ch * 8, kernel_size=3)
self.upscale1 = Upscale(d_ch * 8, d_ch * 4, kernel_size=3)
self.upscale2 = Upscale(d_ch * 4, d_ch * 2, kernel_size=3)
self.res0 = ResidualBlock(d_ch * 8, kernel_size=3)
self.res1 = ResidualBlock(d_ch * 4, kernel_size=3)
self.res2 = ResidualBlock(d_ch * 2, kernel_size=3)
self.out_conv = nn.Conv2D(d_ch * 2,
3,
kernel_size=1,
padding='SAME',
dtype=conv_dtype)
self.upscalem0 = Upscale(in_ch,
d_mask_ch * 8,
kernel_size=3)
self.upscalem1 = Upscale(d_mask_ch * 8,
d_mask_ch * 4,
kernel_size=3)
self.upscalem2 = Upscale(d_mask_ch * 4,
d_mask_ch * 2,
kernel_size=3)
self.out_convm = nn.Conv2D(d_mask_ch * 2,
1,
kernel_size=1,
padding='SAME',
dtype=conv_dtype)
if 'd' in opts:
self.out_conv1 = nn.Conv2D(d_ch * 2,
3,
kernel_size=3,
padding='SAME',
dtype=conv_dtype)
self.out_conv2 = nn.Conv2D(d_ch * 2,
3,
kernel_size=3,
padding='SAME',
dtype=conv_dtype)
self.out_conv3 = nn.Conv2D(d_ch * 2,
3,
kernel_size=3,
padding='SAME',
dtype=conv_dtype)
self.upscalem3 = Upscale(d_mask_ch * 2,
d_mask_ch * 1,
kernel_size=3)
self.out_convm = nn.Conv2D(d_mask_ch * 1,
1,
kernel_size=1,
padding='SAME',
dtype=conv_dtype)
else:
self.out_convm = nn.Conv2D(d_mask_ch * 2,
1,
kernel_size=1,
padding='SAME',
dtype=conv_dtype)
def on_build(self, *args, **kwargs):
self.conv1 = nn.Conv2D(
self.in_ch,
self.out_ch,
kernel_size=self.kernel_size,
strides=2,
padding='SAME',
dilations=self.dilations,
kernel_initializer=conv_kernel_initializer)
def on_build(self, in_planes, out_planes):
self.in_planes = in_planes
self.out_planes = out_planes
self.bn1 = nn.BatchNorm2D(in_planes)
self.conv1 = nn.Conv2D (in_planes, out_planes//2, kernel_size=3, strides=1, padding='SAME', use_bias=False )
self.bn2 = nn.BatchNorm2D(out_planes//2)
self.conv2 = nn.Conv2D (out_planes//2, out_planes//4, kernel_size=3, strides=1, padding='SAME', use_bias=False )
self.bn3 = nn.BatchNorm2D(out_planes//4)
self.conv3 = nn.Conv2D (out_planes//4, out_planes//4, kernel_size=3, strides=1, padding='SAME', use_bias=False )
if self.in_planes != self.out_planes:
self.down_bn1 = nn.BatchNorm2D(in_planes)
self.down_conv1 = nn.Conv2D (in_planes, out_planes, kernel_size=1, strides=1, padding='VALID', use_bias=False )
else:
self.down_bn1 = None
self.down_conv1 = None
def on_build(self, in_ch, d_ch, d_mask_ch, is_hd):
self.is_hd = is_hd
self.upscale0 = Upscale(in_ch, d_ch * 8, kernel_size=3)
self.upscale1 = Upscale(d_ch * 8, d_ch * 4, kernel_size=3)
self.upscale2 = Upscale(d_ch * 4, d_ch * 2, kernel_size=3)
if is_hd:
self.res0 = UpdownResidualBlock(in_ch,
d_ch * 8,
kernel_size=3)
self.res1 = UpdownResidualBlock(d_ch * 8,
d_ch * 4,
kernel_size=3)
self.res2 = UpdownResidualBlock(d_ch * 4,
d_ch * 2,
kernel_size=3)
self.res3 = UpdownResidualBlock(d_ch * 2,
d_ch,
kernel_size=3)
else:
self.res0 = ResidualBlock(d_ch * 8, kernel_size=3)
self.res1 = ResidualBlock(d_ch * 4, kernel_size=3)
self.res2 = ResidualBlock(d_ch * 2, kernel_size=3)
self.out_conv = nn.Conv2D(d_ch * 2,
3,
kernel_size=1,
padding='SAME')
self.upscalem0 = Upscale(in_ch,
d_mask_ch * 8,
kernel_size=3)
self.upscalem1 = Upscale(d_mask_ch * 8,
d_mask_ch * 4,
kernel_size=3)
self.upscalem2 = Upscale(d_mask_ch * 4,
d_mask_ch * 2,
kernel_size=3)
self.out_convm = nn.Conv2D(d_mask_ch * 2,
1,
kernel_size=1,
padding='SAME')
def on_build(self ):
self.upscale0 = Upscale(inter_dims, d_dims*8, kernel_size=3)
self.upscale1 = Upscale(d_dims*8, d_dims*8, kernel_size=3)
self.upscale2 = Upscale(d_dims*8, d_dims*4, kernel_size=3)
self.upscale3 = Upscale(d_dims*4, d_dims*2, kernel_size=3)
self.res0 = ResidualBlock(d_dims*8, kernel_size=3)
self.res1 = ResidualBlock(d_dims*8, kernel_size=3)
self.res2 = ResidualBlock(d_dims*4, kernel_size=3)
self.res3 = ResidualBlock(d_dims*2, kernel_size=3)
self.upscalem0 = Upscale(inter_dims, d_mask_dims*8, kernel_size=3)
self.upscalem1 = Upscale(d_mask_dims*8, d_mask_dims*8, kernel_size=3)
self.upscalem2 = Upscale(d_mask_dims*8, d_mask_dims*4, kernel_size=3)
self.upscalem3 = Upscale(d_mask_dims*4, d_mask_dims*2, kernel_size=3)
self.upscalem4 = Upscale(d_mask_dims*2, d_mask_dims*1, kernel_size=3)
self.out_convm = nn.Conv2D( d_mask_dims*1, 1, kernel_size=1, padding='SAME', dtype=conv_dtype)
self.out_conv = nn.Conv2D( d_dims*2, 3, kernel_size=1, padding='SAME', dtype=conv_dtype)
self.out_conv1 = nn.Conv2D( d_dims*2, 3, kernel_size=3, padding='SAME', dtype=conv_dtype)
self.out_conv2 = nn.Conv2D( d_dims*2, 3, kernel_size=3, padding='SAME', dtype=conv_dtype)
self.out_conv3 = nn.Conv2D( d_dims*2, 3, kernel_size=3, padding='SAME', dtype=conv_dtype)
def on_build(self, in_ch, base_ch, out_ch=None):
self.convs = [ nn.Conv2D( in_ch, base_ch, kernel_size=7, strides=1, padding='SAME'),
nn.Conv2D( base_ch, base_ch, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch, base_ch*2, kernel_size=3, strides=2, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*2, base_ch*2, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*2, base_ch*4, kernel_size=3, strides=2, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*4, base_ch*4, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*4, base_ch*8, kernel_size=3, strides=2, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*8, base_ch*8, kernel_size=3, strides=1, use_bias=False, padding='SAME')
]
self.frns = [ None,
nn.FRNorm2D(base_ch),
nn.FRNorm2D(base_ch*2),
nn.FRNorm2D(base_ch*2),
nn.FRNorm2D(base_ch*4),
nn.FRNorm2D(base_ch*4),
nn.FRNorm2D(base_ch*8),
nn.FRNorm2D(base_ch*8),
]
self.tlus = [ nn.TLU(base_ch),
nn.TLU(base_ch),
nn.TLU(base_ch*2),
nn.TLU(base_ch*2),
nn.TLU(base_ch*4),
nn.TLU(base_ch*4),
nn.TLU(base_ch*8),
nn.TLU(base_ch*8),
]
if out_ch is not None:
self.out_conv = nn.Conv2D( base_ch*8, out_ch, kernel_size=1, strides=1, use_bias=False, padding='VALID')
else:
self.out_conv = None
def on_build(self, in_ch, d_ch, d_mask_ch, n_upscales=4):
self.ups = []
self.res = []
self.upms = []
last_ch = in_ch
last_ch_m = in_ch
kernel_size = 3
for i in reversed(range(1, n_upscales)):
cur_ch = d_ch * (min(2**i, 8))
m_cur_ch = d_mask_ch * (min(2**i, 8))
self.ups.append(
Upscale(last_ch, cur_ch, kernel_size=kernel_size))
self.res.append(ResidualBlock(cur_ch, kernel_size=3))
self.upms.append(
Upscale(last_ch_m,
m_cur_ch,
kernel_size=kernel_size))
last_ch = self.ups[-1].get_out_ch()
last_ch_m = self.upms[-1].get_out_ch()
self.out_conv = nn.Conv2D(d_ch * 2,
3,
kernel_size=1,
padding='SAME')
if 'd' in opts:
self.out_conv1 = nn.Conv2D(d_ch * 2,
3,
kernel_size=3,
padding='SAME')
self.out_conv2 = nn.Conv2D(d_ch * 2,
3,
kernel_size=3,
padding='SAME')
self.out_conv3 = nn.Conv2D(d_ch * 2,
3,
kernel_size=3,
padding='SAME')
self.upscalem3 = Upscale(d_mask_ch * 2,
d_mask_ch * 1,
kernel_size=3)
self.out_convm = nn.Conv2D(d_mask_ch * 1,
1,
kernel_size=1,
padding='SAME')
else:
self.out_convm = nn.Conv2D(d_mask_ch * 2,
1,
kernel_size=1,
padding='SAME')
def on_build(self, lmrks_ch, base_ch, out_ch):
self.convs = [ nn.Conv2D( base_ch*8+lmrks_ch, base_ch*8, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*8, base_ch*8*4, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*8, base_ch*4, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*4, base_ch*4*4, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*4, base_ch*2, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*2, base_ch*2*4, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
nn.Conv2D( base_ch*2, base_ch, kernel_size=3, strides=1, use_bias=False, padding='SAME'),
]
self.frns = [ nn.FRNorm2D(base_ch*8),
nn.FRNorm2D(base_ch*8*4),
nn.FRNorm2D(base_ch*4),
nn.FRNorm2D(base_ch*4*4),
nn.FRNorm2D(base_ch*2),
nn.FRNorm2D(base_ch*2*4),
nn.FRNorm2D(base_ch),
]
self.tlus = [ nn.TLU(base_ch*8),
nn.TLU(base_ch*8*4),
nn.TLU(base_ch*4),
nn.TLU(base_ch*4*4),
nn.TLU(base_ch*2),
nn.TLU(base_ch*2*4),
nn.TLU(base_ch),
]
self.use_upscale = [ False,
True,
False,
True,
False,
True,
False,
]
self.out_conv = nn.Conv2D( base_ch, out_ch, kernel_size=3, strides=1, padding='SAME')
def on_build(self, *args, **kwargs):
self.conv1 = nn.Conv2D(self.in_ch,
self.out_ch * 4,
kernel_size=self.kernel_size,
padding='SAME')
def on_build(self):
self.conv1 = nn.Conv2D(3,
64,
kernel_size=3,
strides=1,
padding='SAME')
self.dense1 = nn.Dense(1, 64, use_bias=False)
self.dense2 = nn.Dense(1, 64, use_bias=False)
self.e0_conv0 = nn.Conv2D(64,
64,
kernel_size=3,
strides=1,
padding='SAME')
self.e0_conv1 = nn.Conv2D(64,
64,
kernel_size=3,
strides=1,
padding='SAME')
self.e1_conv0 = nn.Conv2D(64,
112,
kernel_size=3,
strides=1,
padding='SAME')
self.e1_conv1 = nn.Conv2D(112,
112,
kernel_size=3,
strides=1,
padding='SAME')
self.e2_conv0 = nn.Conv2D(112,
192,
kernel_size=3,
strides=1,
padding='SAME')
self.e2_conv1 = nn.Conv2D(192,
192,
kernel_size=3,
strides=1,
padding='SAME')
self.e3_conv0 = nn.Conv2D(192,
336,
kernel_size=3,
strides=1,
padding='SAME')
self.e3_conv1 = nn.Conv2D(336,
336,
kernel_size=3,
strides=1,
padding='SAME')
self.e4_conv0 = nn.Conv2D(336,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.e4_conv1 = nn.Conv2D(512,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.center_conv0 = nn.Conv2D(512,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.center_conv1 = nn.Conv2D(512,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.center_conv2 = nn.Conv2D(512,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.center_conv3 = nn.Conv2D(512,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.d4_conv0 = nn.Conv2D(1024,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.d4_conv1 = nn.Conv2D(512,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.d3_conv0 = nn.Conv2D(848,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.d3_conv1 = nn.Conv2D(512,
512,
kernel_size=3,
strides=1,
padding='SAME')
self.d2_conv0 = nn.Conv2D(704,
288,
kernel_size=3,
strides=1,
padding='SAME')
self.d2_conv1 = nn.Conv2D(288,
288,
kernel_size=3,
strides=1,
padding='SAME')
self.d1_conv0 = nn.Conv2D(400,
160,
kernel_size=3,
strides=1,
padding='SAME')
self.d1_conv1 = nn.Conv2D(160,
160,
kernel_size=3,
strides=1,
padding='SAME')
self.d0_conv0 = nn.Conv2D(224,
96,
kernel_size=3,
strides=1,
padding='SAME')
self.d0_conv1 = nn.Conv2D(96,
96,
kernel_size=3,
strides=1,
padding='SAME')
self.out1x_conv0 = nn.Conv2D(96,
48,
kernel_size=3,
strides=1,
padding='SAME')
self.out1x_conv1 = nn.Conv2D(48,
3,
kernel_size=3,
strides=1,
padding='SAME')
self.dec2x_conv0 = nn.Conv2D(96,
96,
kernel_size=3,
strides=1,
padding='SAME')
self.dec2x_conv1 = nn.Conv2D(96,
96,
kernel_size=3,
strides=1,
padding='SAME')
self.out2x_conv0 = nn.Conv2D(96,
48,
kernel_size=3,
strides=1,
padding='SAME')
self.out2x_conv1 = nn.Conv2D(48,
3,
kernel_size=3,
strides=1,
padding='SAME')
self.dec4x_conv0 = nn.Conv2D(96,
72,
kernel_size=3,
strides=1,
padding='SAME')
self.dec4x_conv1 = nn.Conv2D(72,
72,
kernel_size=3,
strides=1,
padding='SAME')
self.out4x_conv0 = nn.Conv2D(72,
36,
kernel_size=3,
strides=1,
padding='SAME')
self.out4x_conv1 = nn.Conv2D(36,
3,
kernel_size=3,
strides=1,
padding='SAME')
def on_build(self):
self.minus = tf.constant([104,117,123], dtype=nn.floatx )
self.conv1_1 = nn.Conv2D(3, 64, kernel_size=3, strides=1, padding='SAME')
self.conv1_2 = nn.Conv2D(64, 64, kernel_size=3, strides=1, padding='SAME')
self.conv2_1 = nn.Conv2D(64, 128, kernel_size=3, strides=1, padding='SAME')
self.conv2_2 = nn.Conv2D(128, 128, kernel_size=3, strides=1, padding='SAME')
self.conv3_1 = nn.Conv2D(128, 256, kernel_size=3, strides=1, padding='SAME')
self.conv3_2 = nn.Conv2D(256, 256, kernel_size=3, strides=1, padding='SAME')
self.conv3_3 = nn.Conv2D(256, 256, kernel_size=3, strides=1, padding='SAME')
self.conv4_1 = nn.Conv2D(256, 512, kernel_size=3, strides=1, padding='SAME')
self.conv4_2 = nn.Conv2D(512, 512, kernel_size=3, strides=1, padding='SAME')
self.conv4_3 = nn.Conv2D(512, 512, kernel_size=3, strides=1, padding='SAME')
self.conv5_1 = nn.Conv2D(512, 512, kernel_size=3, strides=1, padding='SAME')
self.conv5_2 = nn.Conv2D(512, 512, kernel_size=3, strides=1, padding='SAME')
self.conv5_3 = nn.Conv2D(512, 512, kernel_size=3, strides=1, padding='SAME')
self.fc6 = nn.Conv2D(512, 1024, kernel_size=3, strides=1, padding=3)
self.fc7 = nn.Conv2D(1024, 1024, kernel_size=1, strides=1, padding='SAME')
self.conv6_1 = nn.Conv2D(1024, 256, kernel_size=1, strides=1, padding='SAME')
self.conv6_2 = nn.Conv2D(256, 512, kernel_size=3, strides=2, padding='SAME')
self.conv7_1 = nn.Conv2D(512, 128, kernel_size=1, strides=1, padding='SAME')
self.conv7_2 = nn.Conv2D(128, 256, kernel_size=3, strides=2, padding='SAME')
self.conv3_3_norm = L2Norm(256)
self.conv4_3_norm = L2Norm(512)
self.conv5_3_norm = L2Norm(512)
self.conv3_3_norm_mbox_conf = nn.Conv2D(256, 4, kernel_size=3, strides=1, padding='SAME')
self.conv3_3_norm_mbox_loc = nn.Conv2D(256, 4, kernel_size=3, strides=1, padding='SAME')
self.conv4_3_norm_mbox_conf = nn.Conv2D(512, 2, kernel_size=3, strides=1, padding='SAME')
self.conv4_3_norm_mbox_loc = nn.Conv2D(512, 4, kernel_size=3, strides=1, padding='SAME')
self.conv5_3_norm_mbox_conf = nn.Conv2D(512, 2, kernel_size=3, strides=1, padding='SAME')
self.conv5_3_norm_mbox_loc = nn.Conv2D(512, 4, kernel_size=3, strides=1, padding='SAME')
self.fc7_mbox_conf = nn.Conv2D(1024, 2, kernel_size=3, strides=1, padding='SAME')
self.fc7_mbox_loc = nn.Conv2D(1024, 4, kernel_size=3, strides=1, padding='SAME')
self.conv6_2_mbox_conf = nn.Conv2D(512, 2, kernel_size=3, strides=1, padding='SAME')
self.conv6_2_mbox_loc = nn.Conv2D(512, 4, kernel_size=3, strides=1, padding='SAME')
self.conv7_2_mbox_conf = nn.Conv2D(256, 2, kernel_size=3, strides=1, padding='SAME')
self.conv7_2_mbox_loc = nn.Conv2D(256, 4, kernel_size=3, strides=1, padding='SAME')
